We believed that the NVIDIA-killer would be some other chip manufacturer. We were very wrong

Yesterday NVIDIA had a stumble in the stock market. The shares lost 7% and then recovered part of the fall. Meanwhile, Google grew by about 4%. Both movements had the same origin: the rumor that Meta is considering using Google’s TPU chips in its data centers in 2027.

Why is it important. During the last few years NVIDIA has managed to dominate imperially the AI ​​chip segment. Its accelerator GPUs made the difference, but although other traditional manufacturers such as AMD tried to follow in its wake, the dominance of the company led by Jensen Huang was spectacular. That could change, and the surprise is that the one who threatens that position is Google.

Google prefers to throw balls out. A Google spokesperson explained on CNBC that “Google Cloud is experiencing accelerated demand for both our custom TPUs and NVIDIA GPUs; we are committed to supporting both, as we have for years.”

But they have been preparing the move for a decade. Sundar Pichai’s company has been working on the development of the Tensor Processing Unit since 2015. They launched the first version in 2018 to take advantage of it in its cloud computing business, but little by little these TPUs have been gaining performance and are now promising alternatives for AI loads, both for training and especially inference, as Ironwood demonstrates.

Anthropic already uses them, Meta could do it. Google has already reached a circular financing agreement with Anthropicto which it will supply its TPUs for data centers that work with its model, Claude. The rumors pointed out by The Information make it feasible that Meta reach a similar agreement with Google and use those chips in its data centers. The difference, of course, is the size of Meta versus Anthropic.

NVIDIA shows off its chest. In a post on It is a message with two faces: on the one hand, congratulations. On the other hand, the declaration of intent.

But you already know what’s coming. The CEO of NVIDIA, Jensen Huang, already warned at the investor conference when presenting results of the current situation: the rivalry with Google’s TPUs is increasing. However, he also insisted that Google remains his client and Gemini – which has just been renewed with a spectacular Gemini 3— can run on NVIDIA technology.

Competition is good. All major technology companies try to avoid dependence on NVIDIA, and almost all of them have their own bets. It’s AMDbut also Intel, Microsoft, amazon and of course the aforementioned Google. But apart from them there are proposals such as those from OpenAI, Broadcom or TSMC that with their XPUs they want to end the reign of NVIDIA.

But CUDA is still a lot of CUDA. The development of own chips is promising, but as AMD knows wellNVIDIA continues to have a spectacular wild card with CUDA, the industry standard development platform for AI solutions. The network effect that this technology has generated it’s going to be hard to beatbut Google certainly has resources to try.

Image | World Economic Forum | Hilel Steinberg

In Xataka | That Qualcomm prepares its own AI chips is good news. Whether it has an opportunity in the market is a very different thing.